Deprecated: Function get_magic_quotes_gpc() is deprecated in /srv/BOINC/live-webcode/html/inc/util.inc on line 640
BRP4U (Raspberry Pi , single DM tasks) feedback thread

WARNING: This website is obsolete! Please follow this link to get to the new Albert@Home website!

BRP4U (Raspberry Pi , single DM tasks) feedback thread

Message boards : Problems and Bug Reports : BRP4U (Raspberry Pi , single DM tasks) feedback thread
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · 4 . . . 5 · Next

AuthorMessage
Profile Bikeman (Heinz-Bernd Eggenstein)
Volunteer moderator
Project administrator
Project developer
Avatar

Send message
Joined: 28 Aug 06
Posts: 1483
Credit: 1,864,017
RAC: 0
Message 112349 - Posted: 7 Feb 2013, 10:40:17 UTC

Dear all,

As some Raspberry Pi owners have already noticed, we are now supporting the Raspberry Pi running Raspbian Linux natively on Albert@Home, that is you can use it without an app_info.xml file or similar manual intervention.

All you need is to install the BOINC client, e.g. from the Raspbian repo via

sudo aptitude install boinc-client


and then attach to the project using the boinccmd command line tool. I haven't tried the Boinc Manager on the Raspi, yet.

We have also set up a special pool of workunits, called BRP4U , which is serving smaller-sized tasks for the Raspis (1/8 th the size delivered to CPUs and GPUs for the regular BRP4 application). Otherwise the tasks are identical.

So far we have seen no problems, please feel free to discuss this version here.

Cheers
HB

ID: 112349 · Report as offensive     Reply Quote
Profile MarkJ

Send message
Joined: 28 Feb 08
Posts: 26
Credit: 28,160
RAC: 0
Message 112350 - Posted: 9 Feb 2013, 6:46:30 UTC
Last modified: 9 Feb 2013, 6:47:37 UTC

I have two pi's on the job over here at the moment. A couple more on order. Taking approx 48 hours a work unit with a "medium" overclock (according to raspi-config). All seems to be working well so far.
ID: 112350 · Report as offensive     Reply Quote
Petr Kriz

Send message
Joined: 16 Feb 08
Posts: 3
Credit: 9,256
RAC: 0
Message 112351 - Posted: 11 Feb 2013, 12:43:32 UTC

Hi, I've already write some feedback in this thread. Is there any new progress in this problem?
The application itself is working properly, I am using same settings as MarkJ (medium 900MHz OC) and 1 WU takes 2 days to compute (same as in MarkJs' case).
Thanks.
ID: 112351 · Report as offensive     Reply Quote
Profile Bernd Machenschalk
Volunteer moderator
Project administrator
Project developer
Avatar

Send message
Joined: 15 Oct 04
Posts: 1956
Credit: 6,218,130
RAC: 0
Message 112352 - Posted: 11 Feb 2013, 13:00:06 UTC - in response to Message 112351.  
Last modified: 11 Feb 2013, 13:19:03 UTC

Is there any new progress in this problem?


The way in which BOINC stores server-assigned credit makes it pretty hard to change it once a workunit has been created. We'll certainly change it for the next charge of WUs, but for the ones already generated on this test project I doubt it is worth a day or so of work.

BM

Edit: got a fresh idea and found it wasn't as hard as I expected. Newly validated BRP4U WUs should be granted 63 now.
ID: 112352 · Report as offensive     Reply Quote
Ray_GTI-R

Send message
Joined: 28 Jan 13
Posts: 2
Credit: 1,697
RAC: 0
Message 112353 - Posted: 13 Feb 2013, 2:15:17 UTC - in response to Message 112352.  
Last modified: 13 Feb 2013, 2:15:59 UTC

Is there any new progress in this problem?


The way in which BOINC stores server-assigned credit makes it pretty hard to change it once a workunit has been created. We'll certainly change it for the next charge of WUs, but for the ones already generated on this test project I doubt it is worth a day or so of work.

BM

Edit: got a fresh idea and found it wasn't as hard as I expected. Newly validated BRP4U WUs should be granted 63 now.


I hope you figure out why http://albert.phys.uwm.edu/workunit.php?wuid=197121 only got 63 credits while http://albert.phys.uwm.edu/workunit.php?wuid=197120 got 500???

BTW on my RPi wuid=197121 took longer (@ 66.66%+/- CPU) vs wuid=197120 (@ 100 CPU). And both computers took longer ... Run Time AND CPU Time.

Precis:- 1/3rd slower (CPU) and 63 vs 500 credits???
ID: 112353 · Report as offensive     Reply Quote
Ray_GTI-R

Send message
Joined: 28 Jan 13
Posts: 2
Credit: 1,697
RAC: 0
Message 112354 - Posted: 14 Feb 2013, 2:15:03 UTC - in response to Message 112353.  
Last modified: 14 Feb 2013, 2:15:47 UTC

Meantime ... Albert is stopped and I'm now running MW@H
ID: 112354 · Report as offensive     Reply Quote
Christoph

Send message
Joined: 16 May 07
Posts: 2
Credit: 30,574
RAC: 0
Message 112355 - Posted: 14 Feb 2013, 23:13:43 UTC - in response to Message 112353.  

I hope you figure out why http://albert.phys.uwm.edu/workunit.php?wuid=197121 only got 63 credits while http://albert.phys.uwm.edu/workunit.php?wuid=197120 got 500???
I don't even ... the discrepancy is the result of a deliberate fix, and some old units still have too much credit. So what? Albert is for testing.

Meantime ... Albert is stopped and I'm now running MW@H
Well, goodbye then.
ID: 112355 · Report as offensive     Reply Quote
Alex

Send message
Joined: 1 Mar 05
Posts: 88
Credit: 398,734
RAC: 0
Message 112391 - Posted: 22 Mar 2013, 22:40:58 UTC

Hi,
I just installed native BOINC on my Nexus. In the app-list is no entry for Albert, there is only one for Einstein and no way to enter an URL.
Is there a way to run this app on my Nexus?
Just to make that clear, I'm a newbe in the Android world.
ID: 112391 · Report as offensive     Reply Quote
Profile Bernd Machenschalk
Volunteer moderator
Project administrator
Project developer
Avatar

Send message
Joined: 15 Oct 04
Posts: 1956
Credit: 6,218,130
RAC: 0
Message 112393 - Posted: 23 Mar 2013, 11:00:25 UTC - in response to Message 112391.  

The current ARM App won't run on Android, an Android App is being worked on an should be available after Easter.

BM
ID: 112393 · Report as offensive     Reply Quote
Alex

Send message
Joined: 1 Mar 05
Posts: 88
Credit: 398,734
RAC: 0
Message 112394 - Posted: 23 Mar 2013, 11:17:56 UTC - in response to Message 112393.  

The current ARM App won't run on Android, an Android App is being worked on an should be available after Easter.

BM


Happy to hear that!
ID: 112394 · Report as offensive     Reply Quote
Alex

Send message
Joined: 1 Mar 05
Posts: 88
Credit: 398,734
RAC: 0
Message 112396 - Posted: 25 Mar 2013, 20:27:45 UTC

Well, in the meantime my Nexus crunches for theSkyNet POGS and I'm happy with the performance. OK, an i5 is 6.711 x faster, but who cares ...

I was looking around, the performance / power relation impressed me. I found this one:
http://www.hardkernel.com/renewal_2011/products/prdt_info.php?g_code=G135341370451&tab_idx=1

Might be good for a summer project ...
ID: 112396 · Report as offensive     Reply Quote
Profile MarkJ

Send message
Joined: 28 Feb 08
Posts: 26
Credit: 28,160
RAC: 0
Message 112404 - Posted: 3 Apr 2013, 7:38:16 UTC

A question if I may. When is it likely this Raspberry Pi app might make it to Einstein?

I have been crunching them for a bit and have been returning what appears to be valid work units (that is they have validated). From looking at my completed tasks I seem to get paired up with a cell processor quite often, but rarely it seems another Pi. Maybe I am the only Pi user apart from HBE?

There was mention over at Einstein of getting another months worth of Arecibo data. Will that effect the rollout plan?
ID: 112404 · Report as offensive     Reply Quote
Christoph

Send message
Joined: 16 May 07
Posts: 2
Credit: 30,574
RAC: 0
Message 112405 - Posted: 4 Apr 2013, 17:06:03 UTC - in response to Message 112404.  

The cell processor is so much better, it gathers up lots of work units. There are other users crunching on Pis: Bikeman (Heinz-Bernd Eggenstein), phil w, emtonsit. There were quite a few more a while ago though, you are correct, maybe those went to other projects that were no longer testing or they didn't think the work units remaining would make much a difference?
ID: 112405 · Report as offensive     Reply Quote
Profile Bernd Machenschalk
Volunteer moderator
Project administrator
Project developer
Avatar

Send message
Joined: 15 Oct 04
Posts: 1956
Credit: 6,218,130
RAC: 0
Message 112407 - Posted: 5 Apr 2013, 4:53:03 UTC - in response to Message 112404.  
Last modified: 5 Apr 2013, 4:56:55 UTC

A question if I may. When is it likely this Raspberry Pi app might make it to Einstein?

I have been crunching them for a bit and have been returning what appears to be valid work units (that is they have validated). From looking at my completed tasks I seem to get paired up with a cell processor quite often, but rarely it seems another Pi. Maybe I am the only Pi user apart from HBE?

There was mention over at Einstein of getting another months worth of Arecibo data. Will that effect the rollout plan?


The main point is that not only the application code but also the workunits of "BRP4U" differ from the current "standard" BRP4 workunits on Albert and Einstein.

For Albert we generated BRP4U WUs manually, but we don't have a workunit generator that would continuously produce BRP4U workunits and won't interfere with the current BRP4 workunit generation.

The rough roadmap is:
1. rollout "BRP5" (Parks Perseus Arm Survey) (and possibly the OpenCL version of FGRP2) on Einstein, to feed the GPUs there
2. limit "BRP4" on Einstein to CPUs
3. switch the "bundled" BRP4 workunits on Einstein to "single-DM" WUs that we are running here at Albert as "BRP4U"

Then we can and will publish the ARM application versions on Einstein, otherwise the tasks will take too long (more than a week on an overclocked Raspberry).

I currently expect this process to take about two months. The additional data from Arecibo just means that we aren't under so much pressure for step 1.

BM
ID: 112407 · Report as offensive     Reply Quote
Alex

Send message
Joined: 1 Mar 05
Posts: 88
Credit: 398,734
RAC: 0
Message 112433 - Posted: 21 Apr 2013, 17:50:33 UTC

Hi BM,

can you pls make clear, which app is for the android world?
I got a message from native boinc half an hour ago that a albert app is available. It is called 'Binary Radio Pulsar Search(single DM) 1.02'
I cannot find this app on the server status page.
There is some enthusiasm here on my side; I'm interested in acceptance, performance aso of the android devices.
ID: 112433 · Report as offensive     Reply Quote
Patrick

Send message
Joined: 13 Mar 13
Posts: 18
Credit: 5,642
RAC: 0
Message 112434 - Posted: 22 Apr 2013, 7:08:34 UTC

I´ve downloaded the application 1.02 with native boinc 7.0.36 on a GT-S5570 with Android 2.3.6.
I had to enable "Run CPU versions of applications for which GPU versions are available" it has no gpu but it´s running.
Hopefully it gives a valid result in a week.
It has a Measured floating point speed of 50,44 million ops/sec and a Measured integer speed of 506.59 million ops/sec.
CPU specification is ARMv6-compatible processor rev 5 (v6l) @600MHz.

Good work.
ID: 112434 · Report as offensive     Reply Quote
Profile Bernd Machenschalk
Volunteer moderator
Project administrator
Project developer
Avatar

Send message
Joined: 15 Oct 04
Posts: 1956
Credit: 6,218,130
RAC: 0
Message 112435 - Posted: 22 Apr 2013, 8:15:52 UTC
Last modified: 22 Apr 2013, 8:26:02 UTC

The current Android App version on Albert is experimental to an extent that we are not even announcing it yet for public testing. It is primarily meant to support the development of the "official" BOINC Client for Android. If that application version works with NativeBOINC, fine then, but we currently neither recommend nor support it.

If you attach to Albert@Home with the latest version of NativeBOINC, you probably get v1.02 of the Android App for "BRP4U (single DM)" (as in the thread title). I would suspect this version to be rather slow (possibly even slower than the 1.01 version), especially compared to the v1.03 NEON version that's also there (but only runs on devices that support NEON). But this would probably need to be installed manually in NativeBOINC, and I have no clue on how to do that, and you will need to do that manually again and again for every new version that we publish.

BM
ID: 112435 · Report as offensive     Reply Quote
Patrick

Send message
Joined: 13 Mar 13
Posts: 18
Credit: 5,642
RAC: 0
Message 112436 - Posted: 22 Apr 2013, 10:35:22 UTC

Thank you for that information and i wish you the best in developing and testing that the application can go up to the live servers.
If you watch here it looks like the arm v6 does not support NEON.
But however the arm v6 mobile phones are older models and as of arm v7 (Cortex A Series) supports it.

http://www.arm.com/products/processors/technologies/neon.php

http://www.arm.com/products/processors/instruction-set-architectures/index.php
ID: 112436 · Report as offensive     Reply Quote
Profile Bikeman (Heinz-Bernd Eggenstein)
Volunteer moderator
Project administrator
Project developer
Avatar

Send message
Joined: 28 Aug 06
Posts: 1483
Credit: 1,864,017
RAC: 0
Message 112437 - Posted: 22 Apr 2013, 12:56:06 UTC - in response to Message 112436.  
Last modified: 22 Apr 2013, 13:02:26 UTC

Hi

Indeed the ARM v6 version doesn't support NEON (on purpose: As you notedNEON was introduced with ARMv7).

We do have an ARMv7 NEON version which is considerably faster than the current ARMv6 app version, and I think we can also speed up even the ARMv6 version by a factor close to 2.

Having said that, we are still at a very early stage of this development and the app versions for armv6 and armv7 will be switched on and off without any notice. Some day you will be able to get them, on others not...so don't be disappointed.

At this point we can also not endorse running our apps on NativeBOINC in the sense that we can not rule out that your phone (especially the battery) might overheat and in fact may be damaged. Our apps rely on thermal control in either Android or the BOINC client. We have not tested this with NativeBOINC. All we know for sure is that some Android devices CAN overheat by running all CPU cores under full load.

Cheers
HB
ID: 112437 · Report as offensive     Reply Quote
Patrick

Send message
Joined: 13 Mar 13
Posts: 18
Credit: 5,642
RAC: 0
Message 112438 - Posted: 22 Apr 2013, 14:03:58 UTC

Ok thanks now i´ve understood that.
To the point with battery overheating, if i go on local computing settings under client management i can choose at what for a temperature the client should do a break and i think that works good at the moment.
Just play around with that value and the task should be paused.
I had a buggy wifi connection so i have set the value "Use max % Processors" to 70%
that fixed it.
The other setting "Use max % cpu time" seems to be buggy in native boinc if i turn
that down a little bit the task starts and stops again and again.But i don´t know for sure if that´s a problem of the task or the nativeboinc program.
ID: 112438 · Report as offensive     Reply Quote
1 · 2 · 3 · 4 . . . 5 · Next

Message boards : Problems and Bug Reports : BRP4U (Raspberry Pi , single DM tasks) feedback thread



This material is based upon work supported by the National Science Foundation (NSF) under Grant PHY-0555655 and by the Max Planck Gesellschaft (MPG). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the investigators and do not necessarily reflect the views of the NSF or the MPG.

Copyright © 2024 Bruce Allen for the LIGO Scientific Collaboration